US20210060771A1

Movatterモバイル変換

Info

Publication number: US20210060771A1
Application number: US17/006,482
Authority: US
Inventors: Stoyan Gaydarov
Original assignee: X Development LLC
Current assignee: Intrinsic Innovation LLC
Priority date: 2019-08-30
Filing date: 2020-08-28
Publication date: 2021-03-04
Also published as: WO2021041907A1

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing robot fault recovery using a process definition graph. One of the methods includes receiving an indication that a robot experienced a fault while performing a schedule of commands generated from an original process definition graph, wherein the original process definition graph comprises a plurality of task nodes that represent a plurality of respective tasks to be performed by one or more robots. A recovery process definition graph is generated, wherein the recovery process definition graph includes a motion node that represents a motion action for the robot to take from a current position to a recovery position. On onsite execution engine executes the recovery process definition graph, thereby causing the robot to perform the motion action to move to the recovery position.

Description

CROSS-REFERENCE TO RELATED APPLICATION

This application is a non-provisional of and claims priority to U.S. Provisional Patent Application number 62/894,616, filed on Aug. 30, 2019, the entire contents which are hereby incorporated by reference.

BACKGROUND

This specification relates to robotics, and more particularly to planning robotic movements.

Robotics planning refers to scheduling the physical movements of robots in order to perform tasks. For example, an industrial robot that builds cars can be programmed to first pick up a car part and then weld the car part onto the frame of the car. Each of these actions can themselves include dozens or hundreds of individual movements by robot motors and actuators.

Robotics planning has traditionally required immense amounts of manual programming in order to meticulously dictate how the robotic components should move in order to accomplish a particular task. Manual programming is tedious, time-consuming, and error prone. In addition, a schedule that is manually generated for one workcell can generally not be used for other workcells. In this specification, a workcell is the physical environment in which a robot will operate. Workcells have particular physical properties, e.g., physical dimensions that impose constraints on how robots can move within the workcell. Thus, a manually programmed schedule for one workcell may be incompatible with a workcell having different physical dimensions.

In many industrial robotics applications, the primary success or failure criteria of a schedule is the time it takes to complete a task. For example, at a welding station in a car assembly line, the time it takes for robots to complete welds on each car is a critical aspect of overall throughput of the factory. When using manual programming, it is often difficult or impossible to predict how long the resulting schedule will take to complete the task.

Workcell failures are a common complication with robot planning. For example, if one of the robots of the workcell fails, the robot can enter a failure condition, e.g., a fault mode. If the fault occurs while robots in the workcell are operating, operation ceases. Typically, a human operator has to manually bring each robot in the work cell to a respective maintenance position for each robot. The operator can then either discard the part that was in progress or restart the process in a recovery mode. In recovery mode, the operator has to manually determine which actions the robots in the workcell had completed prior to the fault. The operator can then use that information to restart the operation from the beginning by instructing the robots in the workcell to perform all the movements of the old plan but to not perform the actions that were successfully performed before the fault. These manual processes are time and labor intensive and drag down productivity. First, the manual operation of bringing each robot to a maintenance position takes the workcell out of operation at least until an operator is available to fix the fault. Second, maneuvering the robot to (or from) the maintenance position while avoiding any collisions with the work pieces or other robots, especially when the robot is partly within a work piece, can be time-consuming and requires meticulous operation by a highly skilled operator. Third, the manual entry of workcell tasks that have been completed is error prone and requires a highly skilled operator who can restart the operation at the correct location.

SUMMARY

This specification describes how a system can generate a schedule for one or more robots using an underconstrained process definition graph. In general the system can iteratively apply transformers in order to repeatedly transform the process definition graph until a schedule satisfying one or more goal criteria is reached. After a fault, the system can use this process to bring each robot in the workcell to a maintenance position, sometimes referred to as a recovery position and also to generate a graph for completing the rest of the actions in the cycle without starting over. However, in the failure condition (e.g., as described above), it is not enough to transform the definition graph. The system has to determine which actions have been completed and store that information together with the information of a position of each robot in the workcell. The system can then use the position information for each robot to generate a process definition graph, e.g., an underconstrained process definition graph, which can be referred to as a recovery process definition graph for bringing the robots to a maintenance position. The system can then, for example, apply transformers as discussed above to the recovery process definition graph until a schedule is generated that when executed moves each robot into a corresponding maintenance position. The recovery process definition graph can include one or more motion nodes that create a path for a particular robot that avoids collisions with physical obstacles of the workcell and/or other robots.

In addition, the system can determine which tasks have been completed prior to the fault, and can generate a resume process definition graph, e.g., an underconstrained process definition graph, for the tasks that have not been completed. The system can again apply transforms to the resume process definition graph until a schedule is generated to complete the rest of the tasks. The resume process definition graph includes one or more motion nodes that cause a robot to move from the recovery position to a position of a first task that has not been completed yet. In some implementations, if a robot has been modified, e.g., with a new tool, the resume process definition graph accounts for that modification.

Particular embodiments of the subject matter described in this specification can be implemented so as to realize one or more of the following advantages. Using transformers to iteratively manipulate a process definition graph to generate a robotics schedule dramatically reduces the amount of manual programming required in order to program robots. The system can generate a fully constrained schedule for an arbitrary number of robots from an initial underconstrained process definition graph. Using a process definition graph makes specifying robot actions more flexible, faster, and less error-prone. In addition, the techniques described below allow for exploring a much larger search space than what could be achieved by mere manual programming. Therefore, the resulting schedules are likely to be faster and more efficient than manually programmed schedules.

In the event of a fault in the workcell, the disclosed system can perform a faster and more efficient response to the fault, without operator intervention. In particular, the system can effectively modify a schedule that is currently being executed to account for the fault without halting the overall process, e.g., by continuing the process using remaining robots while holding the faulty robot at a designated recovery position. Costly downtime of the entire assembly line can therefore be reduced. In addition, the tasks that need to be completed can be automatically completed without having to rerun the whole plan from the start while indicating which tasks have been completed so they are not executed again.

The details of one or more embodiments of the subject matter of this specification are set forth in the accompanying drawings and the description below. Other features, aspects, and advantages of the subject matter will become apparent from the description, the drawings, and the claims.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a diagram that illustrates an example system.

FIG. 2 is a flowchart of an example process for iteratively applying transformers to generate a final process definition graph.

FIG. 3 illustrates an example underconstrained process definition graph.

FIGS. 4A-B illustrate generating transitions for a portion of a process definition graph.

FIG. 5 is a visual illustration of a schedule.

FIG. 6 is a block diagram of exemplary actions to cause a robot to perform one or more motion actions to move to a recovery position.

FIG. 7 illustrates a time of a fault relative to the execution of actions ofFIG. 5.

FIG. 8 illustrates an example of a recovery process definition graph.

FIG. 9 is a block diagram of exemplary actions for causing the workcell to perform tasks of the original process definition graph that have not been performed due to the fault.

FIG. 10 illustrates an example of a resume process definition graph.

FIG. 11 illustrates an example of actions in a resume process definition graph where a robot is replaced by another robot.

Like reference numbers and designations in the various drawings indicate like elements.

DETAILED DESCRIPTION

FIG. 1 is a diagram that illustrates an example system100. The system100 is an example of a system that can implement the techniques described in this specification.

The system100 includes a number of functional components, including aprocess definer110, aplanner120, a collection oftransformers130, auser interface device140, anonsite execution engine150, and arobot interface subsystem160. Each of these components can be implemented as computer programs installed on one or more computers in one or more locations that are coupled to each other through any appropriate communications network, e.g., an intranet or the Internet, or combination of networks.

The system100 also includes aworkcell170 that includesN robots170a-n. The overall goal of theplanner120 and other components of the system100 is to generate, from the underconstrainedprocess definition graph115, a schedule that will be executed by therobots170a-nto accomplish one or more tasks. The resulting schedule can be represented in a variety of ways and may be represented as a near-fully constrained or a fully constrained process definition graph, e.g., the workcell-specificprocess definition graph135. As mentioned above, a common goal metric for such schedules is elapsed time, and thus theplanner120 can aim to generate a schedule that causes therobots170a-nto accomplish the one or more tasks in the shortest possible amount of time.

In this specification, a robot is a machine having a base position, one or more movable components, and a kinematic model that can be used to map desired positions, poses, or both in one coordinate system, e.g., Cartesian coordinates, into commands for physically moving the one or more movable components to the desired positions or poses. In this specification, a tool is a device that is part of and is attached at the end of the kinematic chain of the one or more moveable components of the robot. Example tools include grippers, welding devices, and sanding devices. That is, a robot can include one or more tools.

In this specification, a task is an operation to be performed by a tool. For brevity, when a robot has only one tool, a task can be described as an operation to be performed by the robot as a whole. Example tasks include welding, glue dispensing, part positioning, and surfacing sanding, to name just a few examples. Tasks are generally associated with a type that indicates the tool required to perform the task, as well as a position within a workcell at which the task will be performed.

In this specification, a motion plan is a data structure that provides information for executing an action, which can be a task, a cluster of tasks, or a transition. Motion plans can be fully constrained, meaning that all values for all controllable degrees of freedom for the robot are represented explicitly or implicitly; or underconstrained, meaning that some values for controllable degrees of freedom are unspecified. In some implementations, in order to actually perform an action corresponding to a motion plan, the motion plan must be fully constrained to include all necessary values for all controllable degrees of freedom for the robot. Thus, at some points in the planning processes described in this specification, some motion plans may be underconstrained, but by the time the motion plan is actually executed on a robot, the motion plan can be fully constrained. In some implementations, motion plans represent edges in a task graph between two configuration states for a single robot. Thus, generally there is one task graph per robot. In some implementation, a motion plan can include instructions for the robot to “rest,” i.e., to stay in the current position.

In this specification, a motion swept volume is a region of the space that is occupied by a least a portion of a robot or tool during the entire execution of a motion plan. The motion swept volume can be generated by collision geometry associated with the robot-tool system.

In this specification, a transition is a motion plan that describes a movement to be performed between a start point and an end point. The start point and end point can be represented by poses, locations in a coordinate system, or tasks to be performed. Transitions can be underconstrained by lacking one or more values of one or more respective controllable degrees of freedom (DOF) for a robot. Some transitions represent free motions. In this specification, a free motion is a transition in which none of the degrees of freedom are constrained. For example, a robot motion that simply moves from pose A to pose B without any restriction on how to move between these two poses is a free motion. During the planning process, the DOF variables for a free motion are eventually assigned values, and path planners can use any appropriate values for the motion that do not conflict with the physical constraints of the workcell.

In this specification, a schedule is data that assigns each task to at least one robot. A schedule also specifies, for each robot, a sequence of actions to be performed by the robot. A schedule also includes dependency information, which specifies which actions must not commence until another action is finished. A schedule can specify start times for actions, end times for actions, or both.

To initiate schedule generation, a user can provide aprocess description105 and workcell-specific information107 to aprocess definer110. Theprocess description105 can include a high-level description of the tasks to be completed. The workcell-specific information107 can include data that represents properties of the workcell, including physical dimensions, the locations and dimensions of obstacles or other hardware in the workcell, the type and number ofrobots170a-nin the workcell.

From theprocess description105 and the workcell-specific information, theprocess definer110 can generate an underconstrainedprocess definition graph115. Alternatively or in addition, a user can provide the underconstrainedprocess definition graph115 to theplanner120. In this specification, a process definition graph, or for brevity, a graph, is a directed acyclic graph having constraint nodes and action nodes.

Action nodes represent actions for a robot to perform, which can include nodes representing tasks or clusters of tasks, e.g., as specified in the original process description. Action nodes can also represent transitions that robots can perform, e.g., transitions between tasks or other locations in the workcell.

Constraint nodes represent particular relationships between children nodes that must be preserved in any schedule. In general, constraint nodes can represent existence constraints or time constraints. An existence constraint specifies a constraint on which children can be selected. A time constraint specifies a constraint on the timing among children. For example, a constraint node can have as children two action nodes, and the constraint node can represent a time constraint that a first action represented by a first of the two action nodes must start before a second action represented by a second of the two action nodes.

Being an underconstrained process definition graph means that the graph lacks various kinds of information required to actually drive therobots170a-nto accomplish the tasks. The graph can be underconstrained in a variety of ways. For example, the graph can lack any sense of time, scheduling, or ordering between tasks. Being underconstrained can also mean that various properties of task nodes are partially defined or undefined.

Thus, in some implementations theplanner120 can receive aprocess definition graph115 having nodes representing the tasks to be performed as specified by the process description, but without specifying any ordering between the tasks, without specifying any assignment of any of therobots170a-nto any of the tasks, and without specifying what movements the robots should undertake in order to prepare to perform the tasks.

Theplanner120 can then perform an iterative process to begin solving constraints in the underconstrainedprocess definition graph115. The final output of this process is data representing a schedule, which can be a workcell-specificprocess definition graph135, which, for brevity may also be referred to as a final schedule. The workcell-specificprocess definition graph135 contains enough information to drive therobots170a-nto complete the one or more tasks specified in the original underconstrainedprocess definition graph115. Thus, the workcell-specificprocess definition graph135 will generally specify which robots will be performing which tasks. The workcell-specificprocess definition graph135 can also specify the timing, scheduling, ordering and movement actions to be taken by each robot between tasks. Additionally, the movements specified by the workcell-specificprocess definition graph135 can take into account the physical attributes and obstacles of theworkcell170.

Theonsite execution engine150 receives the workcell-specificprocess definition graph135 and issues commands155 to therobot interface system160 in order to actually drive the movements of the moveable components, e.g., the joints, of therobots170a-n. In some implementations, therobot interface subsystem160 provides a hardware-agnostic interface so that thecommands155 issued byonsite execution engine150 are compatible with multiple different versions of robots. During execution therobot interface subsystem160 can reportexecution data157 back to theonsite execution engine150 so that theonsite execution engine150 can make real-time or near real-time adjustments to the robot movements, e.g., due to local faults or other unanticipated conditions.

In execution, therobots170a-ngenerally continually execute the commands specified explicitly or implicitly by the motion plans to perform the various tasks or transitions of the schedule. The robots can be real-time robots, which means that the robots are programmed to continually execute their commands according to a highly constrained timeline. For example, each robot can expect a command from therobot interface subsystem160 at a particular frequency, e.g., 100 Hz or 1 kHz. If the robot does not receive a command that is expected, the robot can enter a fault mode and stop operating.

In some implementations, theplanner120 and theprocess definer110 are cloud-based systems that are physically removed from a facility that physically houses theworkcell170, while theonsite execution engine150 is local to the facility that physically houses theworkcell150. This arrangement allows theplanner120 to use massive cloud-based computing resources to consider many possibilities for robot schedules, while also allowing for real-time reaction to unanticipated events by theonsite execution engine150.

As stated above, theplanner120 can generate a workcell-specificprocess definition graph135 from the initial underconstrainedprocess definition graph115. To do so, theplanner120 can repeatedly apply a number of transformers from a collection oftransformers130. Each transformer is a stateless function that takes as input an underconstrained process definition graph and resolves variables in the underconstrained process definition graph. As part of this process, a transformer can modify action nodes, constraint nodes, or both, by adding, modifying, or deleting these nodes, and generates as output a modified process definition graph. This process is described in more detail below with reference toFIG. 2.

The transformers to be used can also be specified by a user as atransformer configuration117. In other words, the user can specify which of the collection oftransformers130 are to be used when iteratively modifying the initial underconstrainedprocess definition graph115.

Theplanner120 can also optionally invite users to make transformer selections while generating a final schedule. In other words, theplanner120 can provide a candidate process definition graph123 to auser interface device140. Theuser interface device140 can then present a user interface that allows a user to input a user transformer selection125, which directs theplanner120 to perform the next iteration using atransformer selection127 specified by the user. This interactive process can allow theplanner120 to take into consideration constraints and other real-world considerations that were not or could not be specified as part of the original process description.

FIG. 2 is a flowchart of an example process for iteratively applying transformers to generate a final process definition graph. The process can be implemented by one or more computer programs installed on one or more computers and programmed in accordance with this specification. For example, the process can be performed by theplanner120 shown inFIG. 1. For convenience, the process will be described as being performed by a system of one or more computers.

The system receives an underconstrained process definition graph (210). As described above, a process definition graph is a directed acyclic graph that can include at least two different types of nodes, constraint nodes and action nodes.

FIG. 3 illustrates an example underconstrained process definition graph300. The graph includes aroot node310, two

constraint nodes

322 and324, and six

action nodes

332,334,342,344,346, and348.

Each of the action nodes represents one or more physical actions to be performed by a particular robot. For example, an action node can represent an action to move a robot arm, apply a weld, open a gripper, close a gripper, or any other appropriate action that can be performed by a robot. Action nodes can also be composite actions that include multiple other actions. For example, a PickAndPlace composite action can include a separate pick action and a place action.

Notably, the action nodes need not specify which robot in a workcell will perform each action. The action nodes can be partially defined either by specifying some values, e.g., a duration or a location for each action. The action nodes can also be partially defined by specifying options, e.g., a list or a range of acceptable durations.

TheAllOfInOrder constraint node322 specifies that its children nodes have to be performed in a particular order. In other words, theconstraint node322 introduces a constraint thatAction1, represented by theaction node332, has to occur before any of the other actions, and that Action7, represented by theaction node334, has to occur after any of the other actions.

TheAllOf constraint node322 specifies simply that all of its children have to be performed, but does not specify an ordering. Thus, through the iterative planning process, the planner is free to schedule Actions3-6 in any appropriate order that does not violate any other constraints in the graph.

One concrete example of an application that might use a graph similar to the graph300 is welding a part in a factory. For example, theinitial action node332 can represent an action that moves the part into place, and thefinal action node334 can represent an action to move the part to its destination. Those are examples of actions that have to happen first and last, as specified by theAllOfInOrder constraint node322.

Actions3-6 can be welding actions for four separate weld points on the part. For most welding applications, the order does not matter. Thus, while theAllOf constraint node324 imposes a constraint that all welds have to be performed, theAllOf constraint node324 does not impose any constraints on the ordering in which the welds must be performed.

Instead, the planner will use various transformers in order to search the space of schedules that satisfy the constraints imposed by the final graph and choose a schedule that is best according to a particular objective function. For example, the planner can simply search the space of schedules that satisfy the constraints and identify which schedule executes all the actions the fastest.

TABLE 1 lists some common constraint node types.

TABLE 1

Constraint Node Name	Constraint

AllOf	All children must be performed
AnyOf	Any children can be performed
OneOf	Exactly one child must be performed
InOrder	Children must be performed in an order specified
	by the representation in the graph
DirectlyInOrder	Children must be performed in an order specified
	by the representation in the graph, without any
	intervening actions
MustOccurDuring	Child must be performed during a specified time
	period
MustNotOccurDuring	Child must not be performed during a specific
	time period
Conflict	Children cannot occur at overlapping times
Cluster	All children must be performed by the same robot
If	If one child occurs, the subsequent child must
	happen after it

A constraint node can combine multiple constraint node types. For example, in this notation, the AllOfInOrder node had a name that specified two constraint types, both the AllOf constraint type as well as the InOrder constraint type.

As shown inFIG. 2, the system applies local transformers (320). As described above, each transformer is a function that further constrains the process definition by assigning values to variables or manipulating or expanding nodes in the graph.

In general a system can apply a transformer by matching certain patterns associated with nodes in the graph. As one example, applying a transformer can include looking for underconstrained properties in the graph and assigning values to those properties. As another example, a transformer can have one or more defined node types as input and the system can determine that the transformer is a match for the node types when a matching arrangement of nodes found in the graph.

The system can distinguish between local transformers and global transformers and apply local transformers first. A local transformer is a transformer that does not require reconsideration of all nodes in the graph and which affects only a sufficiently small subset of the graph. The system can use any appropriate definition for what is a sufficiently small subset of the graph, e.g., only transformers that affect their direct children or nodes not more than N links away. As a particular example, a local transformer can be applied to a single node in the graph.

An example of a global transformer is a “clustering” transformer, which takes consideration all tasks in the graph that change the position of one or more robots, and proposes ordering constraints that ensure that the robots move efficiently between tasks (for example, avoiding doubling back by robots where possible).

The system can apply local transformers first in order to quickly generate many additional constraints in the graph. For example, if an action node has a particular pose, the system can apply an inverse kinematics (IK) transformer that will generate the kinematic parameters for achieving the pose. Thus, for a given pose, the system can consider an IK transformer to be a local transformer because the transformer affects only a single action node.

The system generates transitions (330). Transitions are actions taken by robots to move from one configuration to another. Thus, if two actions are to be performed by a robot in sequence, the system can generate a transition between the actions by determining how the robot can move from a pose for one action to a pose for another. In some implementations, the system implements transition generators as transformers that seek to match on two action nodes in sequence that so far have no assigned intermediate transition.

Some transformers are designed to generate many alternative options that can all be considered when performing scheduling. For example, when generating transitions, the system can generate multiple different ways of moving from one action to another. The system can represent each generated alternative transition as a separate node in the graph. Since only one transition is needed, the system can constrain the alternatives with an appropriate OneOf constraint node.

FIGS. 4A-B illustrate generating transitions for aportion400aof a process definition graph. InFIG. 4A, anAllOfInOrder constraint node410 represents that a welding action represented by anaction node420 should be performed before a welding action represented by anotheraction node430. In some implementations, during transition generation the system searches the graph for gaps between sequenced actions that lack transitions. Thus, the system can identify theportion400aas a portion of the graph having two sequenced actions but lacking a transition between them.

FIG. 4B illustrates a modifiedportion400bof the process definition graph after transition generation. As shown, the system generated four alternative transitions for transitioning between theaction node420 and theaction node430. Each alternative transition is represented by its own action node, e.g., the

transition nodes

450,460,470, and480. The transition nodes450-480 are constrained by aOneOf constraint node440 that represents that the system should execute one and only one of the generated alternative transitions.

This example illustrates how transformers can add additional constraints to the graph in the form of alternatives rather than selections. In other words, the transformers seek to increase the space of possible alternatives schedules rather than attempting to solve the constraints of a schedule in one pass.

Therefore, after the system performs transition generation, the process definition graph typically has many more nodes that it previously did, with each node representing a possible transition between actions.

As shown inFIG. 2, the system performs conflict identification (240). In other words, the system identifies actions in the graph that cannot occur at the same time, the same space, or both. For example, the system can generate a swept volume for each action node in the graph and identify which actions are potentially conflicting. Notably, the system can perform the conflict identification process before scheduling occurs. The system can then perform a deconfliction process after scheduling occurs. In other words, the system identifies conflicts before scheduling, but need not strive to avoid conflicts during scheduling.

The system performs scheduling (250). In general, the scheduling process solves for underconstrained values in the graph until either no more values need to be solved or the system determines that no solution can be found. As described above, a schedule specifies one robot to perform each task, and for each robot, a sequence of actions to be performed. A schedule can also specify a start time for each action and dependencies between actions.

For example, to perform scheduling, the system can receive as input a set of possible graphs, e.g., a set of every possible alternative graph that can be generated from the graph and that accomplishes the required tasks. The system can process the set of possible graphs to generate an output that is a selection of action nodes to execute as well as dependencies (i.e., a selection of constraint nodes) that specify requirements of the schedule. That is, the output of the scheduling process can be an updated graph, generated from the original graph, with extra constraints that remove the uncertainties of the original graph.

As a particular example, referring to the example depicted inFIG. 4B, the output of the scheduling process might be a graph that includes an AllOfInOrder node with three

child nodes

420,460, and430. That is, the system selected thesecond transition460 from the set of

possible transitions

450,460,470, and480, removing the uncertainty of theOneOf node440.

As another particular example, there may be a situation where two robots need to cross each other, the system can identify (e.g., during conflict identification) the constraint that the two robots cannot collide when crossing each other. The output of the scheduling process can therefore include a scheduling constraint that ensures that the two robots do not perform the movement at the same time, for example, by generating a dependency between the completion of the first movement and the beginning of the second movement (e.g., using an InOrder node or a MustNotOccurDuring node).

In general, the system can use a variety of solvers to determine that no more constraints need to be solved. For example, the system can use a circuit solver to determine when sequenced starting and ending points have no gaps in between. If gaps remain, the system can use transition generation to fill the gaps or raise an error if no solution exists.

The system can also insert rests into the actions and adjust the values of the rests in order to find a scheduling solution. For example, one perhaps not ideal schedule is that a robot can perform an action and then wait for all other robots to finish their tasks before doing anything else. But by using rests, the system increases the chances that a scheduling solution can be found.

As part of the scheduling process, the system can assign tasks to individual robots in the workcell. Thus, at the end of the scheduling process, the system can generate a visual representation of which tasks are performed by which robots and when.

FIG. 5 is a visual illustration of a schedule. The information illustrated inFIG. 5 can still be represented as a process definition graph. And in fact, the deconfliction process that occurs after scheduling can operate on the process definition graph generated by the scheduling process. However, the visualization shown inFIG. 5 is useful for understanding the constraints encoded into the graph.

InFIG. 5, three robotic components in a workcell have been assigned to perform a variety of tasks. Time generally moves from left to right inFIG. 5, and the arrows represent time ordered constraints between actions by different robotic components. For actions performed by the same robotic component, the ordering of the actions implies a time ordered constraint.

Thus, the example schedule illustrated inFIG. 5 first has thefixturing510 of the workcell perform apart transfer512 and then a close clampsaction514. For example, the close clampsaction514 can represent clamps closing down on the part to secure it for welding.

At some time after thetransfer512 starts, thefirst robot520 performs ajoint move522 to get into position for welding. However, theaction weld action524 cannot happen until the actual clamps are closed, as illustrated by the arrow from theclosed clamps action514 to theweld action524.

At some time after the clamps are closed, asecond robot530 performs ajoint move532 to get into position for aweld534. Thejoint move532 also has a constraint that it has to happen after the clamps are closed by the close clampsaction514.

Thefirst robot520 can then perform a secondjoint move526 to get into position for thesecond weld528, but not until the second robot has finished itsjoint move532, as represented by the arrow from thejoint move532 to thejoint move526. The first robot can then perform thesecond weld528.

Finally, thefixturing510 can open the clamps with anopen clamps action516, but not until both

welds

528 and534 are completed. Thefixturing510 can then perform atransfer action518 to move the part along in an assembly line.

As shown inFIG. 2, the system performs deconfliction (260). In general, the deconfliction process also uses transformers that aim to solve possible conflicts in the schedule. As described above, although conflicts may have been identified before scheduling, the scheduler was not necessarily bound by such conflicts during scheduling. In general, generating an initial solution using an underconstrained graph that ignores some conflicts provides computational advantages over trying to solve all constraints and all possible conflicts in one pass.

The system optionally runs additional user-selected transformers (270). As described above, the system can provide a user interface that seamlessly allows some human design in the process. In particular, the system can present a user interface that allows a user to manually specify a next transformer to be applied in the graph or some other manipulation of the graph. As part of this user interface, the system can also present a graphical simulation of the robots executing a particular candidate schedule. For example, if a particular transition between actions seems too awkward or otherwise not ideal, a human can manually select a different transition. As another example, humans tend to be better at performing geometric or spatial groupings. Thus, a human may want to impose a Cluster constraint node for a particular group of actions that are close together in space, time, or both.

The system determines whether a goal has been reached (280). The system can use a goal solver to determine whether a process definition graph meets various goal criteria. As mentioned above, total time is often a critical goal criterion. Each action in the schedule can be associated with a respective duration, and then the system can use the respective durations to determine whether a schedule exists that meets the goal criteria. The system can also use other goal criteria, for example, power used, or some combined score of multiple goal criteria.

If the goal is reached (280), the system outputs the final plan (branch to285). For example, the final plan can then be provided to an onsite execution engine for execution by robots in a workcell.

If the goal is not reached, the system determines whether to perform more iterations (290). In some implementations, the system automatically performs another iteration as long as the system has not determined that a solution is not possible. The system can alternatively or in addition perform an additional iteration only if a maximum number of iterations has not been reached.

If no more iterations are to be performed and the goal has not been reached, the system can raise an error (branch to295). At that point, the users can determine whether to modify the original process definition in order to try to find a valid solution with different inputs.

In some implementations, as depicted inFIG. 2, if more iterations are to be performed, the system returns to step220 and reapplies local transformers. In some other implementations, the system can return to step240 to perform another round of conflict identification. That is, the system might not need to reapply the local transformers or to generate transitions again.

One of the advantages to representing all phases of the planning process as a process definition graph is that the history of the process is represented very explicitly. The state of the schedule as it evolved over the iterations becomes quite clear. This allows actions to be easily undone and the schedule's modifications rewound in time and tried again with different transformers or parameters. This also makes debugging the schedules much easier because it becomes clear how the schedule got to where it is and by which transformers.

The graph modification process described above can be used to cause robots in a workcell to dynamically respond to a fault in the workcell and to address the fault by completing the rest of the tasks in the originally generated process definition graph.

In some implementations, the system can perform the process depicted inFIG. 2 multiple times to generate multiple different candidate schedules. For example, the system might perform the process multiple times in parallel, e.g., by selecting different sequences of transformations to generate the different candidate schedules. The system can then select the final schedule from the set of candidate schedules according to one or more criteria, e.g., a time to complete the one or more required tasks, a measure of efficiency, a measure of safety, etc. That is, the system can search the space of possible schedules by evaluating different transformation sequences. As a particular example, the system might evaluate a particular sequence of transformation sequences and determine to “backtrack” to a particular transformation in the sequence and begin a new sub-sequence from that point.

FIG. 6 is a flowchart of an example process for generating a process definition graph to handle a fault in a workcell. The process can be implemented by one or more computer programs installed on one or more computers in one or more locations and programmed in accordance with this specification. For example, the process can be performed by theplanner120 shown inFIG. 1. For convenience, the process will be described as being performed by a system of one or more computers.

The system receives an indication that a robot experienced a fault while performing a schedule of commands generated from an original process definition graph (610). The original process definition graph refers to a graph with multiple task nodes that represent respective tasks to be performed by one or more robots. For example, the original process definition plan can be afinal plan285 shown inFIG. 2 or a workcell-specificprocess definition graph135 ofFIG. 1.

FIG. 7 illustrates the location of a fault during execution of an example schedule.FIG. 7 includes afault line710 that illustrates a time of a fault relative to the execution of actions illustrated inFIG. 5. When a fault is detected, theonsite execution engine150 or another component within the system100 can detect the fault and save the execution state of the system. For example, theonsite execution engine150 can save a state of each task. As toFIG. 7, the save state can include completed tasknodes including transfer512,close clamps514,joint move522,joint move532 and,weld action524. The save state can also include action nodes that have not been completed, which can include the following action nodes:joint move526,weld action528, weld action,open clamps action516, andtransfer518. The save state also tracks the dependencies of actions. For example, inFIG. 7 the save state includes a dependency from close clamps forweld524 andjoint move532 as well as the other dependencies shown including a dependency betweenjoint move532 andjoint move526 and a dependency betweenopen claims action516 and each ofweld action528 andweld action534. The save state can include other data as appropriate.

FIG. 8 illustrates an example of a recovery process definition graph. Theroot node810 of the graph represents the transition to maintenance position process. The basic structure of the recovery process definition graph is a transition falling between a fault position and a recovery position. Thus, the system can generate anAllOfInOrder constraint node822 representing the sequencing of these steps of the process. The fault and recovery positions of the robot can be represented as respective action nodes in the graph. And although there may be no tasks to be performed at these locations, the system can treat these nodes as task nodes. This means that the normal graph modification processes will automatically generate one or more possible transitions between these positions.

Thus, theAllOfInOrder constraint node822 represents that thefault position action832 occurs before therecovery position action834. Thefault position action832 can include a motion plan that modifies a tool of the robot that makes it safer to move the robot holding the tool to the recovery position. For example, the motion plan can retract a welding tip or disable normal functionality of the tool, e.g., disabling power to a welding tip. Similarly, therecovery position action834 can include a motion plan that is related to a tool safety for a tool mounted on the robot, which can make it safer for human operators to tend to the robot at the recovery position.

In some implementationfault position action832 andrecovery position action834 are not explicitly represented in the graph. Instead, the system can generate transitions from implicit representations of the fault position and the recovery position. TheOneOf constraint node824 indicates that exactly one of the candidate transitions842,844,846, or848 should be performed to move the robot to the recovery position. The system can generate a similar graph for each robot, taking into account possible collisions in order to move each robot into its respective recovery position.

As shown inFIG. 6, the system provides the recovery process definition graph to an onsite execute engine, e.g., onsite execution engine150 (630).

The onsite execution engine can then execute the recovery process definition graph (640), thereby causing the robot to perform the transition between the fault position and the recovery position. The execution of the recovery process definition graph can proceed in a similar manner as that of any other process definition graph.

Notably, the onsite execution engine may execute the recovery process definition graph exactly once. That is, when the recovery process definition graph is executed, the positions of the robots have changed in a way that executing the graph again is not desirable or extendable to other situations in the workcell.

When each robot moves to its designated recovery position, the system can recover and complete the tasks that are left to finish after the fault, and without requiring the robots to perform any of the movement actions or tasks that were already completed before the fault. Instead, in some cases, the system can generate a resume process definition graph that can also be executed just once before the system resumes execution of the original process definition graph. In some other cases, the system can also generate a resume process definition graph to be immediately executed only by the remaining, operable robots in the workcell, i.e., while the faulty robot remains at the designated recovery position. This can reduce costly downtime of an entire assembly line.

FIG. 9 is a flowchart an example process for generating a process definition graph to perform actions that have not been performed due to the fault. The process can be implemented by one or more computer programs installed on one or more computers in one or more locations and programmed in accordance with this specification. For example, the process can be performed by theplanner120 shown inFIG. 1. For convenience, the process will be described as being performed by a system of one or more computers.

The system generates a resume process definition graph (910). The resume process definition graph includes a subset of the task nodes in the original process definition graph. The system can generate the resume process definition graph by using the save state described above to determine which actions still need to be performed and what the dependencies are for those actions. For example, as shown inFIG. 7 the save state can include data that indicates that when a fault occurred

tasks

526,528,534,516, and518 have not been completed yet. The save state can also include dependencies for those tasks as shown inFIG. 7. And based on this portion of the save state data the system can generate a resume process definition graph.

However, there is no requirement that the resume process definition graph result in a schedule that assigns the same robots to perform the same tasks. Rather, the system can modify the initial resume process definition graph per its normal transformation pipeline, which can result in increased efficiency for the tasks that remain. For example, when a fault causes a single robot to malfunction and fail to complete two of its tasks, the resume process definition graph can result in a modified schedule in which the two remaining tasks are performed concurrently by two robots. The two robots may or may not include the robot that was originally assigned to perform the task.

FIG. 10 illustrates an example of a resume process definition graph. The root node1010 of the graph represents the resume process for uncompleted tasks. TheAllOfInOrder constraint node1022 has three children,node1024,node1034, andnode1036, withnode1024 having two child nodes, e.g.,node1044 and node1046). These nodes correspond to the actions needed to complete the tasks illustrated inFIG. 7. TheAllOf constraint node1024 indicates that bothWeld21044 andWeld31046 have to be completed regardless of the order. However, based on the graph,

actions

1044 and1046 have to be completed prior to theopen clamps action1034 andtransfer action1036. In some implementations, the system can reuse some of the nodes of the original process definition graph. For example, if the original process definition graph had

action

1034 and1036 as children of an AllOfinOrder node, the system can reuse that portion of the graph. However, as stated above, the system can also generate the resume process definition graph from an underconstrained initial process definition graph in the same way that a normal process definition graph would be generated for all the tasks of the workcell.

The system can also take into account, when building the resume process definition graph, the reason for the fault. For example, if the fault was due to a robot breaking down, the system can take that into account and remove the robot from performing any task. This, naturally, can result in a different assignment of robots to tasks for both the resume process as well as all subsequent workcell iterations going forward.

FIG. 11 illustrates an example schedule generated from a resume process definition graph. In this example, a robot is replaced by another robot. InFIG. 11

Robot

1520 (not shown) has been replaced byRobot3521. The actions have also been updated, because, for example, different joint moves are now needed to perform the two welds. Thus, joint move532 (not shown) is now replaced byjoint move533 andjoint move526 is now replaced byjoint move527. The open clampsaction516 is still dependent on bothweld528 andweld534 being completed.

The system can also add other nodes, if needed, to build the resume process definition graph. For example, the resume process definition graph can include a motion node that represents a motion action that causes the robot to move from the recovery position to a position to a first task in the subset of tasks represented in the original process definition graph. Thus, the system detects that the robots are in the recovery position and builds the resume process definition graph accordingly.

As shown inFIG. 9, the system provides the resume process definition graph to the onsite execution engine (920). In other words, the system can generate a schedule using the resume process definition graph, which the onsite execution engine can use to drive the robots in the workcell. This process can be executed in the same manner as described in relation toFIG. 1. Thus, the system, e.g.,planner120, can reside on a remote system while the onsite execution engine, e.g.,onsite execution engine150, can be local to the workcell, e.g.,workcell170.

This process can also be used to account for faults that result in alterations of physical characteristics of robots. For example, if the fault was due to a tool breaking and the tool gets replaced by a new tool, the system can generate a resume process definition graph that takes into account the altered physical characteristics of the new tool. In addition, the system can generate a new schedule from the original process definition graph that also takes into consideration the altered physical characteristics of the new tool. The system can then use the new schedule for all subsequent workcell iterations. The system can then provide to the onsite execution engine the modified process definition graph and the onsite execution engine can execute the modified process definition graph for all iterations after executing the resume process definition graph once.

The robot functionalities described in this specification can be implemented by a hardware-agnostic software stack, or, for brevity just a software stack, that is at least partially hardware-agnostic. In other words, the software stack can accept as input commands generated by the planning processes described above without requiring the commands to relate specifically to a particular model of robot or to a particular robotic component. For example, the software stack can be implemented at least partially by theonsite execution engine150 and therobot interface subsystem160 ofFIG. 1.

The software stack can include multiple levels of increasing hardware specificity in one direction and increasing software abstraction in the other direction. At the lowest level of the software stack are robot components that include devices that carry out low-level actions and sensors that report low-level statuses. For example, robots can include a variety of low-level components including motors, encoders, cameras, drivers, grippers, application-specific sensors, linear or rotary position sensors, and other peripheral devices. As one example, a motor can receive a command indicating an amount of torque that should be applied. In response to receiving the command, the motor can report a current position of a joint of the robot, e.g., using an encoder, to a higher level of the software stack.

Each next highest level in the software stack can implement an interface that supports multiple different underlying implementations. In general, each interface between levels provides status messages from the lower level to the upper level and provides commands from the upper level to the lower level.

Typically, the commands and status messages are generated cyclically during each control cycle, e.g., one status message and one command per control cycle. Lower levels of the software stack generally have tighter real-time requirements than higher levels of the software stack. At the lowest levels of the software stack, for example, the control cycle can have actual real-time requirements. In this specification, real-time means that a command received at one level of the software stack must be executed and optionally, that a status message be provided back to an upper level of the software stack, within a particular control cycle time. If this real-time requirement is not met, the robot can be configured to enter a fault state, e.g., by freezing all operation.

At a next-highest level, the software stack can include actuator feedback controllers. An actuator feedback controller can include control logic for controlling multiple robot components through their respective motor feedback controllers. For example, some robot components, e.g., a joint arm, can actually be controlled by multiple motors. Thus, the actuator feedback controller can provide a software abstraction of the joint arm by using its control logic to send commands to the motor feedback controllers of the multiple motors.

At a next-highest level, the software stack can include joint feedback controllers. A joint feedback controller can represent a joint that maps to a logical degree of freedom in a robot. Thus, for example, while a wrist of a robot might be controlled by a complicated network of actuators, a joint feedback controller can abstract away that complexity and expose that degree of freedom as a single joint. Thus, each joint feedback controller can control an arbitrarily complex network of actuator feedback controllers. As an example, a six degree-of-freedom robot can be controlled by six different joint feedback controllers that each control a separate network of actual feedback controllers.

Each level of the software stack can also perform enforcement of level-specific constraints. For example, if a particular torque value received by an actuator feedback controller is outside of an acceptable range, the actuator feedback controller can either modify it to be within range or enter a fault state.

To drive the input to the joint feedback controllers, the software stack can use a command vector that includes command parameters for each component in the lower levels, e.g. a position, torque, and velocity, for each motor in the system. To expose status from the joint feedback controllers, the software stack can use a status vector that includes status information for each component in the lower levels, e.g., a position, velocity, and torque for each motor in the system. In some implementations, the command vectors also include some limited information regarding constraints to be enforced by the controllers in the lower levels.

At a next-highest level, the software stack can include joint collection controllers. A joint collection controller can handle issuing of command and status vectors that are exposed as a set of part abstractions. Each part can include a kinematic model, e.g., for performing inverse kinematic calculations, limit information, as well as a joint status vector and a joint command vector. For example, a single joint collection controller can be used to apply different sets of policies to different subsystems in the lower levels. The joint collection controller can effectively decouple the relationship between how the motors are physically represented and how control policies are associated with those parts. Thus, for example if a robot arm has a movable base, a joint collection controller can be used to enforce a set of limit policies on how the arm moves and to enforce a different set of limit policies on how the movable base can move.

At a next-highest level, the software stack can include joint selection controllers. A joint selection controller can be responsible for dynamically selecting between commands being issued from different sources. In other words, a joint selection controller can receive multiple commands during a control cycle and select one of the multiple commands to be executed during the control cycle. The ability to dynamically select from multiple commands during a real-time control cycle allows greatly increased flexibility in control over conventional robot control systems.

At a next-highest level, the software stack can include joint position controllers. A joint position controller can receive goal parameters and dynamically compute commands required to achieve the goal parameters. For example, a joint position controller can receive a position goal and can compute a set point for achieving the goal.

At a next-highest level, the software stack can include Cartesian position controllers and Cartesian selection controllers. A Cartesian position controller can receive as input goals in Cartesian space and use inverse kinematics solvers to compute an output in joint position space. The Cartesian selection controller can then enforce limit policies on the results computed by the Cartesian position controllers before passing the computed results in joint position space to a joint position controller in the next lowest level of the stack. For example, a Cartesian position controller can be given three separate goal states in Cartesian coordinates x, y, and z. For some degrees, the goal state could be a position, while for other degrees, the goal state could be a desired velocity.

These functionalities afforded by the software stack thus provide wide flexibility for control directives to be easily expressed as goal states in a way that meshes naturally with the higher-level planning techniques described above. In other words, when the planning process uses a process definition graph to generate concrete actions to be taken, the actions need not be specified in low-level commands for individual robotic components. Rather, they can be expressed as high-level goals that are accepted by the software stack that get translated through the various levels until finally becoming low-level commands. Moreover, the actions generated through the planning process can be specified in Cartesian space in a way that makes them understandable for human operators, which makes debugging and analyzing the schedules easier, faster, and more intuitive. In addition, the actions generated through the planning process need not be tightly coupled to any particular robot model or low-level command format. Instead, the same actions generated during the planning process can actually be executed by different robot models so long as they support the same degrees of freedom and the appropriate control levels have been implemented in the software stack.

Embodiments of the subject matter and the functional operations described in this specification can be implemented in digital electronic circuitry, in tangibly-embodied computer software or firmware, in computer hardware, including the structures disclosed in this specification and their structural equivalents, or in combinations of one or more of them. Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible non-transitory storage medium for execution by, or to control the operation of, data processing apparatus. The computer storage medium can be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them. Alternatively or in addition, the program instructions can be encoded on an artificially-generated propagated signal, e.g., a machine-generated electrical, optical, or electromagnetic signal, that is generated to encode information for transmission to suitable receiver apparatus for execution by a data processing apparatus.

The term “data processing apparatus” refers to data processing hardware and encompasses all kinds of apparatus, devices, and machines for processing data, including by way of example a programmable processor, a computer, or multiple processors or computers. The apparatus can also be, or further include, special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit). The apparatus can optionally include, in addition to hardware, code that creates an execution environment for computer programs, e.g., code that constitutes processor firmware, a protocol stack, a database management system, an operating system, or a combination of one or more of them.

A computer program which may also be referred to or described as a program, software, a software application, an app, a module, a software module, a script, or code) can be written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages, and it can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment. A program may, but need not, correspond to a file in a file system. A program can be stored in a portion of a file that holds other programs or data, e.g., one or more scripts stored in a markup language document, in a single file dedicated to the program in question, or in multiple coordinated files, e.g., files that store one or more modules, sub-programs, or portions of code. A computer program can be deployed to be executed on one computer or on multiple computers that are located at one site or distributed across multiple sites and interconnected by a data communication network.

For a system of one or more computers to be configured to perform particular operations or actions means that the system has installed on it software, firmware, hardware, or a combination of them that in operation cause the system to perform the operations or actions. For one or more computer programs to be configured to perform particular operations or actions means that the one or more programs include instructions that, when executed by data processing apparatus, cause the apparatus to perform the operations or actions.

As used in this specification, an “engine,” or “software engine,” refers to a software implemented input/output system that provides an output that is different from the input. An engine can be an encoded block of functionality, such as a library, a platform, a software development kit (“SDK”), or an object. Each engine can be implemented on any appropriate type of computing device, e.g., servers, mobile phones, tablet computers, notebook computers, music players, e-book readers, laptop or desktop computers, PDAs, smart phones, or other stationary or portable devices, that includes one or more processors and computer readable media. Additionally, two or more of the engines may be implemented on the same computing device, or on different computing devices.

The processes and logic flows described in this specification can be performed by one or more programmable computers executing one or more computer programs to perform functions by operating on input data and generating output. The processes and logic flows can also be performed by special purpose logic circuitry, e.g., an FPGA or an ASIC, or by a combination of special purpose logic circuitry and one or more programmed computers.

Computers suitable for the execution of a computer program can be based on general or special purpose microprocessors or both, or any other kind of central processing unit. Generally, a central processing unit will receive instructions and data from a read-only memory or a random access memory or both. The essential elements of a computer are a central processing unit for performing or executing instructions and one or more memory devices for storing instructions and data. The central processing unit and the memory can be supplemented by, or incorporated in, special purpose logic circuitry. Generally, a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto-optical disks, or optical disks. However, a computer need not have such devices. Moreover, a computer can be embedded in another device, e.g., a mobile telephone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a Global Positioning System (GPS) receiver, or a portable storage device, e.g., a universal serial bus (USB) flash drive, to name just a few.

Computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media and memory devices, including by way of example semiconductor memory devices, e.g., EPROM, EEPROM, and flash memory devices; magnetic disks, e.g., internal hard disks or removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks.

To provide for interaction with a user, embodiments of the subject matter described in this specification can be implemented on a computer having a display device, e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor, for displaying information to the user and a keyboard and pointing device, e.g, a mouse, trackball, or a presence sensitive display or other surface by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback, e.g., visual feedback, auditory feedback, or tactile feedback; and input from the user can be received in any form, including acoustic, speech, or tactile input. In addition, a computer can interact with a user by sending documents to and receiving documents from a device that is used by the user; for example, by sending web pages to a web browser on a user's device in response to requests received from the web browser. Also, a computer can interact with a user by sending text messages or other forms of message to a personal device, e.g., a smartphone, running a messaging application, and receiving responsive messages from the user in return.

Embodiments of the subject matter described in this specification can be implemented in a computing system that includes a back-end component, e.g., as a data server, or that includes a middleware component, e.g., an application server, or that includes a front-end component, e.g., a client computer having a graphical user interface, a web browser, or an app through which a user can interact with an implementation of the subject matter described in this specification, or any combination of one or more such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication, e.g., a communication network. Examples of communication networks include a local area network (LAN) and a wide area network (WAN), e.g., the Internet.

The computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. In some embodiments, a server transmits data, e.g., an HTML page, to a user device, e.g., for purposes of displaying data to and receiving user input from a user interacting with the device, which acts as a client. Data generated at the user device, e.g., a result of the user interaction, can be received at the server from the device.

While this specification contains many specific implementation details, these should not be construed as limitations on the scope of any invention or on the scope of what may be claimed, but rather as descriptions of features that may be specific to particular embodiments of particular inventions. Certain features that are described in this specification in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination. Moreover, although features may be described above as acting in certain combinations and even initially be claimed as such, one or more features from a claimed combination can in some cases be excised from the combination, and the claimed combination may be directed to a subcombination or variation of a subcombination.

Particular embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. For example, the actions recited in the claims can be performed in a different order and still achieve desirable results. As one example, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In certain some cases, multitasking and parallel processing may be advantageous.